The US FDA clears the first serology test for assessing infectious spread
In many ways it is difficult to assess the passage of time while in a national shutdown. It was a little over a week ago that I wrote about the need for a good antibody test for SARS-CoV-2, in order to get decent tracking of the extent of the infection.
I wrote then the distinction between rRT-PCR tests and antibody tests to measure immune response. By having a widespread asseessment of how many individuals have developed antibodies against SARS-CoV-2, there is a path to identify people who are immune to becoming infected, and thus can serve a vital role in the current phase of the epidemic in the US as immune plasma donors for blood treatment of infection. In addition these immune individuals can serve vital roles where infection risk is high (whether working in a grocery store or a delivery person).
Last week I was hopeful that the Abbott Laboratories application was around their antibody platform iStat, however it turned out to be the ID Now real-time platform which is a 5- to 15-minute point-of-care test (POCT) platform with over 18,000 installations in doctor’s offices and clinics nationwide. While it will help assess the presence of viral RNA in the throat and nasal cavities, what is becoming clear to me is two things.
First is that our rRT-PCR tests are sensitive. Here’s the FDA Emergency Use Authorization site, accessing the LabCorp documentation of an EUA Summary PDF indicates the level of stringent validation they went through, showing sensitivity of the test down to 6.5 RNA molecules. You can dig around and see numbers a little higher in the 10 molecule or 15 molecule range.
Second however, is that rRT-PCR tests still may not pick up a clearly diseased COVID-19 positive patient. When I first read this piece in the New York Times titled “If you have coronavirus symptoms, assume you have the test, even if you test negative”, I was suspicous of the anectdotal stories of a 30% false-negative rate until I started digging into this. And there was this pre-print from China (linked to in the editorial), then this one I located.
Paper after paper, it looked like there were problems with the test itself. However with the stated validation numbers, how could rRT-PCR come up with so many false-negatives? Then I find this paper titled “The comparative superiority of IgM-IgG antibody test to real-time reverse transcriptase PCR detection for SARS-CoV-2 infection diagnosis”, and this one in JAMA whose figure I’ve placed here.
Focusing on the blue dots, and the 40 Ct threshold (dotted line), you can see how nasal swabs from 20 patients in the hospital with COVID-19 are very high – and very low. Indeed of the 8 patients tested via nasal swabs (admittedly a very low number) they only had a 63% positive rate (5 out of 8). This is not a fault of the testing, nor do I believe it to be a fault of sample handling; I believe this is the nature of the course of the disease. The highest success rate (93%, 14/15 samples) was bronchoalveolar lavage fluid.
Thus the New York Times physician from Yale is correct: if you have coronavirus symptoms, assume you have the illness, even if you test negative.
The exciting news however is the headline: the FDA did approve its first antibody test (also called a serology test) against the protective IgG and IgM antibodies from individuals exposed to the virus. While hard data is still being generated on the number of infected individuals without symptoms, one prominent AAAS/Science study put the pre-symptomatic and asymptomatic number at 55%. As I mentioned before (and glad there is increasing journalistic attention paid to this, for example here and here),
Take a close look at the above graph, however. It breaks down the asymptomatic at 30%, the immune response at around day 5 to day 10 post-infection (that is of the 85% who have no or mild symptoms). It is these individuals that an antibody test is especially valuable.
Take a close look at the above graphic too. This is tracking mild cases (that 55%) and shows the interplay between IgM and IgG immunoglobulin levels, their natural rise and fall, and their longevity. This is important to understand in the value of having an accurate SARS-CoV-2 IgG and IgM test available.
Last time I wrote up a list of five companies saying publicly they were applying for an Emergency Use Authorization, and Cellex US was not one of them. The authorization letter (available on the FDA EUA page) was dated April 1, and yesterday was the first I saw it mentioned.
The test is lateral-flow, and takes only 15 minutes from a small amount of plasma, serum or blood, and is very simple to perform. The graphic below explains the process – no swabs or preserving fluid or purification columns or real-time PCR instruments, as with rRT-PCR. And it needs to be done in a CLIA medium or high complexity laboratory. Here is the Instructions for Use (IFU) document (PDF).
Here is a table of their accuracy; the sensitivity is 93.75% and the specificity is 96.40% – this may cause problems with false-positives when a large number of negative individuals are tested.
As of this writing 3 April 2020) there are 245,601 confirmed cases in the US. Assuming another 30% pre- or asymptomatic, that is about 319,000 cases in a country with 1000 times the population (329 million), or a prevalence of 0.1%. You need to test 10 thousand random individuals to pick up 3 individuals with asymptomatic or presymptomatic infection. With a specificity of 96.40%, of 10,000 individuals that’s 360 people who are false-positives.
But that is what other diagnostic tools are for; the rRT-PCR molecular test and CT scans are the main tools, I am sure there are others.
This is a huge step forward to look at the scope of the disease in the US and a key component of the path forward. On that note, Scott Gottlieb (former FDA commissioner) put together this National Coronavirus Response – a Roadmap to Reopening, where Phase II surveillance and careful re-opening would include knowledge of the recovered and resistant per an antibody test that is widely available and widely disseminated. This first antibody EUA is a great advance in that direction.
> [Cellex test] sensitivity is 93.75% and the specificity is 96.40%
It is unexpectedly good news that the performance is so high. When I looked at the Cellex test’s package insert, the table differed slightly from the screenshot in the original post. However, the take-home message is the same — the test is sensitive and specific.
The revised? table on page 3 still lists the result of tests on 378 patients. Significantly, the positive comparators are defined as 128 “RT-PCR positive clinical specimens.” 98 are “collected from individuals who tested positive [by RT-PCR] and were quarantined in a makeshift hospital.” Another 30 “were collected from hospitalized individuals who were [RT-PCR positive] and exhibited severe symptoms.” 120 of these 128 were positive by the Cellex test. Of the 250 negatives, 10 incorrectly tested positive by the Cellex test.
So the specificity is slightly worse than shown in the blue-tinted table above.
One key question, unanswered here, is how many of those people with Covid-19 disease who test negative by RT-PCR, are (correctly) reported as “positives” by the Cellex test. As you say, about 30% of RT-PCR tests return negative results for patients known to have the disease. The Wang et al figure showing viral RNA levels on eight different types of sample indicates that it’s not due to a flaw in the RT-PCR tests. Rather, it looks like some patients shed so little virus that even sensitive tests can’t detect it.
While the Cellex immunoassay can’t be positive prior to IgM production at about day 7 post-infection: how good is it at identifying RT-PCR negative patients after that point, who have Covid-19 disease?
The second point that the label doesn’t address is whether the Cellex test could be used for screening purposes. This set of 378 patients can’t answer that question. We would need to see a second study, comparing asymptomatic and/or mildly-symptomatic Covid-19 patents against a comparable group of people without a SARS-CoV-2 infection. It’s easy to see that the Cellex test’s sensitivity would be much lower in this setting. Specificity might stay at 96%, or be lower.
Here’s the arithmetic assuming that the performance is 60%/96% for a general asymptomatic/mild-symptoms population, if 2% of the population is infected but with mild or no symptoms. For 1,000 people tested, 12 of the 20 actually-infected people would get a positive result. However, 39 of the 980 actually-uninfected people would also get a positive result. If one goal of screening is “more efficient use of health-system resources,” this might not help. (For contact tracing: very useful.)